Complex Mechanical Properties of Steel

Whereas considerable progress has been reported on the quantitative estimation of the microstructure of steels as a function of most of the important determining variables, it remains the case that it is impossible to calculate all but the simplest of mechanical properties given a comprehensive description of the structure at all conceivable scales. Properties which are important but fall into this category are impact toughness, fatigue, creep and combinations of these phenomena. The work presented in this thesis is an attempt to progress in this area of complex mechanical properties in the context of steels, although the outcomes may be more widely applied. The approach used relies on the creation of physically meaningful models based on the neural network and genetic programming techniques. It appears that the hot–strength, of ferritic steels used in the power plant industry, diminishes in concert with the dependence of solid solution strengthening on temperature, until a critical temperature is reached where it is believed that climb processes begin to contribute. It is demonstrated that in this latter regime, the slope of the hot–strength versus temperature plot is identical to that of creep rupture–strength versus temperature. This significant outcome can help dramatically reduce the requirement for expensive creep testing. Similarly, a model created to estimate the fatigue crack growth rates for a wide range of ferritic and austenitic steels on the basis of static mechanical data has the remarkable outcome that it applies without modification to nickel based superalloys and titanium alloys. It has therefore been possible to estimate blindly the fatigue performance of alloys whose chemical composition is not known. Residual stress is a very complex phenomenon especially in bearings due to the Hertzian contact which takes place. A model has been developed that is able to quantify the residual stress distribution, under the raceway of martensitic ball bearings, using the running conditions. It is evident that a well–formulated neural network model can not only be extrapolated even beyond material type, but can reveal physical relationships which are found to be informative and useful in practice.

Name dc:type.qualificationname

Doctor of Philosophy (PhD)

Level dc:type.qualificationlevel

Doctoral

Grantor dc:publisher.institution

University of Cambridge

Year dc:date.issued

2009

Author dc:creator

• Dimitriu, Radu

Licence

• https://apollo8-f-pro.lib.cam.ac.uk/bitstreams/7364518d-fd62-429e-a936-c81559386fd4/download
• https://www.rioxx.net/licenses/all-rights-reserved/

Language dc:language

en

DOI dc:identifier.doi

https://doi.org/10.17863/CAM.14246

OAI identifier oai:identifier

oai:www.repository.cam.ac.uk:1810/218319